CN113312411A - Equipment fault cause diagnosis method based on knowledge graph and rule constraint - Google Patents

Equipment fault cause diagnosis method based on knowledge graph and rule constraint Download PDF

Info

Publication number
CN113312411A
CN113312411A CN202110692310.4A CN202110692310A CN113312411A CN 113312411 A CN113312411 A CN 113312411A CN 202110692310 A CN202110692310 A CN 202110692310A CN 113312411 A CN113312411 A CN 113312411A
Authority
CN
China
Prior art keywords
fault
equipment
diagnosis
rule
reason
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110692310.4A
Other languages
Chinese (zh)
Other versions
CN113312411B (en
Inventor
张谞
荆巍巍
倪菁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
CETC 14 Research Institute
Original Assignee
CETC 14 Research Institute
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by CETC 14 Research Institute filed Critical CETC 14 Research Institute
Priority to CN202110692310.4A priority Critical patent/CN113312411B/en
Publication of CN113312411A publication Critical patent/CN113312411A/en
Application granted granted Critical
Publication of CN113312411B publication Critical patent/CN113312411B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2465—Query processing support for facilitating data mining operations in structured databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • G06N5/025—Extracting rules from data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/048—Fuzzy inferencing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Fuzzy Systems (AREA)
  • Databases & Information Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Automation & Control Theory (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Test And Diagnosis Of Digital Computers (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses an equipment fault cause diagnosis method based on knowledge graph and rule constraint, which combines equipment fault knowledge graph and fault diagnosis rule constraint set to carry out intelligent diagnosis of given equipment fault. The method can realize the reason diagnosis of the given fault in a dynamic self-adaptive mode on the basis of the equipment fault knowledge map and the fault diagnosis rule constraint set.

Description

Equipment fault cause diagnosis method based on knowledge graph and rule constraint
Technical Field
The invention relates to equipment fault diagnosis technology, in particular to an equipment fault reason diagnosis method based on knowledge graph and rule constraint.
Background
With the development of knowledge graph technology, a large amount of experience knowledge and unstructured data in the field of equipment can be reorganized and fully utilized, and become important knowledge resources for equipment development and test guarantee. Meanwhile, the traditional maintenance support means and mode are time-consuming and labor-consuming, especially the fault diagnosis of complex equipment depends on expert ability and experience, the rapid and accurate diagnosis scene of the equipment cannot be met, and the intelligent equipment fault diagnosis requirement is increasingly highlighted. The artificial intelligence technology based on the knowledge graph provides brand-new technical support for the field of equipment, particularly for fault reason positioning in the process of equipment fault diagnosis. The intelligent fault diagnosis based on the knowledge map not only needs artificial intelligence technologies such as natural language processing and the like to extract knowledge of the unstructured fault diagnosis text, but also needs to construct accurate diagnosis rules by combining fault diagnosis results. Therefore, by combining with actual requirements of equipment guarantee, a complete equipment fault knowledge map and diagnosis rule set are constructed, and an intelligent diagnosis method is designed, so that a new idea can be provided for timely diagnosing and positioning equipment faults under an information condition.
In the prior art, fault diagnosis methods based on knowledge bases in chinese patents with patent numbers CN108520093B and CN109189866A are both fault diagnosis methods for example level and dependency similarity calculation, and mainly include: constructing a knowledge base of an equipment instance hierarchy; collecting equipment fault states and characteristics; and performing similarity calculation according to the stored characteristic frequencies of the fault characteristic frequency knowledge base to realize fault diagnosis.
The limitations of the above patents are:
and the fault cause diagnosis is carried out only by relying on similarity calculation, so that the self-adaption is poor and the interpretability is low.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides the equipment fault cause diagnosis method based on knowledge graph and rule constraint, which is based on equipment fault knowledge graph and fault diagnosis rule constraint set, can realize cause diagnosis of given fault in a dynamic self-adaptive mode and can effectively analyze cause-effect relation between equipment fault phenomenon and fault cause.
The purpose of the invention is realized by the following technical scheme.
A method for diagnosing equipment fault causes based on knowledge graph and rule constraint comprises the following steps:
s1, constructing an equipment fault knowledge graph data model: defining basic elements of equipment fault diagnosis, wherein the basic elements of the equipment fault diagnosis comprise different dimensions of fault phenomena in the equipment fault diagnosis, fault reason elements and semantic relations between the phenomena and reasons; the RDF data model is used for expressing and organizing different dimensions of fault phenomena, fault reasons and semantic relations among the fault phenomena;
s2, constructing an equipment fault knowledge graph: giving a specific device, extracting key elements in fault description, mining the dependency relationship between fault phenomena and reasons, generating a device fault knowledge graph from a fault diagnosis text of an open domain, and representing and storing the device fault knowledge graph by using an RDF (remote data format) graph;
s3, mining the rule from the fault phenomenon set to the fault reason to generate a diagnosis rule constraint set of the given fault;
and S4, based on the fault phenomenon of the given current equipment, defining a fault reason intelligent selection technology based on rule matching and knowledge reasoning by combining the equipment knowledge graph constructed in the S2 and the rule set constructed in the S3, screening out a possible fault reason list, and generating a final fault reason.
Further, step S1 is specifically as follows:
s101, representing equipment fault knowledge graph as binary group
Figure BDA0003126611280000021
Wherein
Figure BDA0003126611280000022
Describing equipment and troubleshooting in a knowledge graph representing equipment failuresThe data pattern of the basic elements in the process of interruption, also called the equipment fault ontology,
Figure BDA0003126611280000023
the entity graph is composed of concrete faults, reason factors and phenomena in the equipment fault knowledge graph and semantic relations among the faults, the reason factors and the phenomena;
s102, for equipment fault body
Figure BDA0003126611280000024
Further expressed as doublets
Figure BDA0003126611280000025
Wherein the content of the first and second substances,
Figure BDA0003126611280000026
is a set related to related concepts in the process of equipment and fault diagnosis, and comprises five major classes of equipment names, fault phenomena, fault elements and fault reasons, wherein each major class comprises a plurality of subclasses,
Figure BDA0003126611280000027
is a collection of semantic relationships between concepts, including inheritance, dependency, causal relationships;
s103, knowledge graph entity graph for equipment fault
Figure BDA0003126611280000028
Can be represented as a doublet
Figure BDA0003126611280000029
Wherein epsilon represents a specific entity corresponding to the equipment fault ontology concept, and comprises a specific equipment, a specific fault phenomenon, a specific fault element and a specific fault reason;
Figure BDA00031266112800000210
is a collection of semantic relationships between equipment entities, where
Figure BDA00031266112800000211
S104, for a given equipment fault scene to be diagnosed, an equipment entity e in the scene belongs to epsilon, and the concept category corresponding to the entity is positioned to an equipment fault body
Figure BDA00031266112800000212
A node of
Figure BDA00031266112800000213
Further, step S2 is specifically as follows:
s201, collecting diagnosis description texts related to equipment faults by taking sentences as units, and expressing the diagnosis description texts as
Figure BDA00031266112800000214
Where n represents the total number of sentences present in the troubleshooting text sequence, using a segmentation tool pair
Figure BDA00031266112800000215
Performing word segmentation to obtain a fault phrase sequence of the equipment
Figure BDA00031266112800000216
n represents the total number of fault phrases, then equipment fault diagnosis experts are organized to label each phrase, and equipment entities are labeled, wherein the equipment entities comprise equipment names, fault phenomena, fault elements and fault reasons, and otherwise, the equipment entities are labeled as non-equipment entities;
s202, defining field characteristics of equipment fault phrases, designing a characteristic selection module, and generating a characteristic vector p corresponding to each phraseiObtaining a feature expression of the whole fault diagnosis phrase sequence
Figure BDA0003126611280000031
Wherein d represents a feature vector dimension for each fault phrase;
s203, giving labeled phrases and feature vectors, wherein each sentence is formed by labels labeled by equipment failure diagnosis expertsWhether a phrase is a set of tags for an equipment entity, denoted as
Figure BDA0003126611280000032
Wherein liWhether the phrase is an equipment entity is represented, namely an equipment name, a fault phenomenon, a fault element, a fault reason and a non-equipment entity;
s204, using a classification algorithm to obtain
Figure BDA0003126611280000033
And
Figure BDA0003126611280000034
the discriminant function is obtained from the middle learning
Figure BDA0003126611280000035
It satisfies the minimization function:
Figure BDA0003126611280000036
s205, for the residual texts of the fault diagnosis texts, extracting equipment fault entities of the whole fault diagnosis texts by using the discriminant function;
s206, constructing a relation set between the equipment fault entities based on the semantic relation set defined in the S103;
and S207, representing the equipment fault entity set obtained by mining, the equipment fault entity semantic relation set and the equipment fault body as an RDF (remote data format) graph for storage, and finally generating an equipment fault knowledge graph.
Further, the feature selection mode of the feature selection module in step S202 specifically includes: the equipment fault diagnosis description text specific grammatical features, fault related vocabulary statistical features, phrase semantic features of the equipment fault diagnosis text and phrase structure features of the equipment fault diagnosis text, wherein the fault related vocabulary statistical features comprise field specific words, naming rules and digital text combination specifications, and the phrase semantic features of the equipment fault diagnosis text comprise phrase vector sequences generated by using a pre-training model Bert.
Further, step S3 is specifically as follows:
s301, based on given equipment fault knowledge graph
Figure BDA0003126611280000037
Collecting the existing related fault entities completing fault diagnosis and semantic relations to form a fault diagnosis subgraph;
s302, designing a failure diagnosis causal relationship learning module, and finding a causal relationship between equipment, a failure phenomenon and a failure reason from semantic relationships between different equipment failure entities: firstly, obtaining training data of causal relationship learning from an existing fault-fault phenomenon-fault cause set, wherein a feedforward neural network is adopted to encode the state of each equipment entity in a fault diagnosis subgraph, the relationship between two states is encoded into a real-value matrix form, matrix elements are not fixed {0, 1} binary codes, but attention scores are introduced to serve as confidence degrees of rules to represent the possibility of causal relationship between the entities; secondly, training an LSTM network model, and carrying out modeling representation on the relation and the state of each step;
s303, designing a logic rule generation module, and learning a rule expression for fault reason diagnosis in a fault diagnosis subgraph: firstly, in order to reduce the noise influence of irrelevant fault entities, an attention mechanism is constructed on a fault entity feature vector, a corresponding fault type is judged for a vector with higher weight, further, essential key feature elements for primary fault occurrence are obtained, a rule generator is designed on the basis, and a rule causing the fault occurrence is mined.
Further, in step S303, the content of the fault diagnosis rule generator is as follows:
determining a fault entity and a causal relationship set under a given fault, performing feature extraction and logic constraint on key elements related to the fault to form complete logic Rule description, introducing a fuzzy factor representing inference reliability on the basis of the definition of an A → B causal relationship generating formula, and obtaining a Rule constraint fuzzy generating formula in the fault diagnosis process, wherein the Rule constraint fuzzy generating formula is expressed as (RC, RS and mu), and the RC represents a Rule constraint condition set; RS represents a fault diagnosis result set corresponding to the rule constraint condition set; the fuzzy factor mu represents the credibility of the rule, and the generating formula is triggered when the causal relationship occurs to the related fault entity in the fault diagnosis process, and the final rule set is formed through induction.
Further, step S4 is specifically as follows:
s401, designing a fault reason intelligent selection technology based on matching and knowledge reasoning based on the equipment knowledge graph constructed in S2 and the rule set constructed in S3, and for this reason, firstly designing a fault phenomenon and fault reason matching method based on fault diagnosis rules:
Figure BDA0003126611280000041
where s (u, upsilon) represents the similarity between fault phenomena, lυ，iAnd lυ，iMatching degrees of the fault phenomena upsilon to the fault reasons i and j are respectively obtained, then, a random walk model is adopted to train the size relation of the matching degrees of the fault reasons i and j, the matching degrees of other similar fault phenomena u to the fault reasons i and j are obtained, based on the process, updating of the matching degree evaluation data of the fault phenomena is iteratively completed, and a plurality of candidate fault reasons are obtained.
S401, after the fault reason rule matching is completed, a plurality of candidate diagnosis reasons are obtained, the final diagnosis reasoning is converted into a knowledge reasoning problem based on supervised learning, the thought of RankSVM is adopted, fault equipment, fault phenomena and key elements are comprehensively considered, the partial order relation among the fault reasons is used as a feature vector for training, and then the fault reason sequencing is converted into a comparative classification problem among the reasons, wherein the formula is defined as follows:
Figure BDA0003126611280000051
yi(w·(xi-xi))≥1-ξi
wherein the content of the first and second substances,
Figure BDA0003126611280000052
and
Figure BDA0003126611280000053
respectively, the failure cause xiThe scores in the fault features σ and υ, w is the weight vector that needs to be adjusted step by step in the learning process, and the parameter C identifies the degree of tradeoff between model complexity and training error, ξiAnd training the data based on the optimization function to generate a sequencing model, further finishing sequencing a given fault reason set, and finally selecting the fault reason with the highest rank as a diagnosis result of the current fault.
Compared with the prior art, the invention has the advantages that:
1. the equipment names, the fault phenomena, the fault elements and the fault reasons are used for jointly constructing the knowledge graph, so that the equipment faults can be more accurately modeled and analyzed.
2. By using a semi-supervised algorithm, equipment elements and semantic relations in the unstructured text can be automatically extracted, and the time cost of building the map by relying on an artificial mode too much is relieved.
3. The rule for mining the fault phenomenon set to the fault reason can provide basis and interpretability for the standard positioning of equipment faults;
4. based on the fault phenomenon of the given current equipment, the equipment knowledge graph and the rule constraint set are combined, and intelligent diagnosis is carried out by adopting rule matching and knowledge reasoning, so that the self-adaption and robustness of the diagnosis process are ensured.
In conclusion, the method can define the equipment name, the fault phenomenon, the fault element and the fault reason in a fine-grained manner, construct the equipment fault index map and the rule constraint set, and then intelligently select and position the fault reason based on rule matching and knowledge reasoning, and the comprehensive algorithm can provide more accurate and reliable service for intelligent diagnosis of the equipment fault.
Drawings
FIG. 1 is a block diagram of the method of the present invention;
FIG. 2 is a flow chart of the equipment fault knowledge graph construction of the present invention.
Detailed Description
The invention is described in detail below with reference to the drawings and specific examples.
Referring to fig. 1, the equipment fault cause diagnosis method based on knowledge graph and rule constraint of the invention comprises the following steps:
s1, constructing an equipment fault knowledge map data model, and defining basic elements of equipment fault diagnosis, wherein the basic elements mainly comprise different dimensions of fault phenomena, fault reason elements and semantic relations between the phenomena and reasons in the equipment fault diagnosis; the RDF data model is used for expressing and organizing different dimensions of fault phenomena, fault reasons and semantic relations among the fault phenomena, and the method comprises the following specific steps:
s101, the equipment fault knowledge graph can be expressed as a binary group
Figure BDA0003126611280000061
Wherein
Figure BDA0003126611280000062
The data mode which represents the basic elements in the equipment fault knowledge graph and describes the equipment and the fault diagnosis process is also called as the equipment fault body,
Figure BDA0003126611280000063
and the entity graph is formed by representing specific faults, reason factors and phenomena in the equipment fault knowledge graph and semantic relations among the faults, the reason factors and the phenomena.
S102, for equipment fault body
Figure BDA0003126611280000064
Can be further represented as a binary group
Figure BDA0003126611280000065
Wherein，
Figure BDA0003126611280000066
The method is a set related to related concepts in the process of equipment and fault diagnosis, and mainly comprises five major classes of equipment names, fault phenomena, fault elements and fault reasons, wherein each major class comprises a plurality of subclasses.
Figure BDA0003126611280000067
Is a collection of semantic relationships (inheritance, dependency, cause and effect) between concepts.
S103, knowledge graph entity graph for equipment fault
Figure BDA0003126611280000068
Can be represented as a doublet
Figure BDA0003126611280000069
Wherein epsilon represents a specific entity corresponding to the equipment fault ontology concept, and mainly represents a specific equipment, a specific fault phenomenon, a specific fault element and a specific fault reason.
Figure BDA00031266112800000610
Is a collection of semantic relationships between equipment entities, where
Figure BDA00031266112800000611
S104, for a given equipment fault scene to be diagnosed, an equipment entity e in the scene belongs to epsilon, and the concept category corresponding to the entity can be positioned to an equipment fault body
Figure BDA00031266112800000612
A node of
Figure BDA00031266112800000613
S2, as shown in fig. 2, constructing an equipment fault knowledge graph, providing a specific equipment, extracting key elements in the fault description, mining the dependency between the fault phenomenon and the cause, generating an equipment fault knowledge graph, and representing and storing the equipment fault knowledge graph by using an RDF graph, the specific steps are as follows:
s201, collecting diagnosis description texts related to equipment faults by taking sentences as units, and expressing as S ═ S1，...，si，...，snN, where n represents the total number of sentences present in the troubleshooting text sequence. Using word segmentation tool pairs
Figure BDA00031266112800000614
Performing word segmentation to obtain a fault phrase sequence of the equipment
Figure BDA00031266112800000615
m represents the total number of failure phrases. And then organizing equipment fault diagnosis experts to label each phrase, and marking equipment entities such as equipment names, fault phenomena, fault elements, fault reasons and the like, and vice versa (labeled as non-equipment entities).
S202, defining field characteristics of equipment fault phrases, designing a characteristic selection module, and generating a characteristic vector p corresponding to each phraseiObtaining a feature expression of the whole fault diagnosis phrase sequence
Figure BDA00031266112800000616
Where d represents the feature vector dimension for each fault phrase. The feature selection mode of the feature selection module is specifically as follows:
{ specific grammatical features of equipment fault diagnosis description text, fault-related vocabulary statistical features (domain-specific words, naming rules, digital text combination specifications, etc.), phrase semantic features of equipment fault diagnosis text (phrase vector sequence generated by using pre-training model Bert), and phrase structural features of equipment fault diagnosis text }
S203, giving labeled phrases and feature vectors, wherein labels labeled by equipment fault diagnosis experts in each sentence form a label set whether the phrase is an equipment entity or not, and the label set is represented as
Figure BDA0003126611280000071
Wherein liIndicating whether the phrase is an equipment entity, i.e., equipment name, failure phenomenon, failure element, failure cause, non-equipment entity.
S204, using a classification algorithm (random forest, support vector machine and the like) to obtain
Figure BDA0003126611280000072
And
Figure BDA0003126611280000073
the discriminant function is obtained from the middle learning
Figure BDA0003126611280000074
It satisfies the minimization function:
Figure BDA0003126611280000075
and S205, extracting equipment fault entities of the whole fault diagnosis text by using the discriminant function for the residual texts of the fault diagnosis text.
And S206, constructing a relation set between the equipment fault entities based on the semantic relation set defined in the S103.
And S207, representing the equipment fault entity set obtained by mining, the equipment fault entity semantic relation set and the equipment fault body as an RDF (remote data format) graph for storage, and finally generating an equipment fault knowledge graph.
S3, mining the rule from the fault phenomenon set to the fault reason, and generating the diagnosis rule constraint set of the given fault, the concrete steps are as follows:
s301, based on given equipment fault knowledge graph
Figure BDA0003126611280000076
And collecting the existing related fault entities and semantic relations for completing fault diagnosis to form a fault diagnosis subgraph.
S302, designing a failure diagnosis causal relationship learning module, and finding a causal relationship between equipment, a failure phenomenon and a failure reason from semantic relationships between different equipment failure entities. Firstly, training data for causal relationship learning are obtained from an existing fault-fault phenomenon-fault cause set. The method comprises the steps that states of all equipment entities in a feedforward neural network coding fault diagnosis subgraph are adopted, the relation between the two states is coded in a real-value matrix form, matrix elements are not fixed {0, 1} binary codes, attention scores are introduced to serve as regular confidence degrees, and the probability of causal relation between the entities is expressed; secondly, training an LSTM network model, and modeling and representing the relation and the state of each step.
S303, designing a logic rule generating module, and learning a rule expression for fault reason diagnosis in the fault diagnosis subgraph. Firstly, in order to reduce the noise influence of irrelevant fault entities, an attention mechanism can be constructed on the fault entity feature vectors, corresponding fault categories are judged for the vectors with higher weights, further, essential key feature elements for primary fault occurrence are obtained, a rule generator is designed on the basis, and the rules causing the fault occurrence are mined. The content of the fault diagnosis rule generator is as follows:
and determining a fault entity and a causal relationship set under a given fault, and performing feature extraction and logic constraint on key elements related to the fault to form a complete logic rule description. On the basis of the definition of the A → B causal relationship generating formula, a fuzzy factor representing inference reliability can be introduced to obtain a Rule constraint fuzzy generating formula in the fault diagnosis process, which can be expressed as < Rule > ═ (RC, RS, mu), wherein RC represents a Rule constraint condition set; RS represents a fault diagnosis result set corresponding to the rule constraint condition set; the ambiguity factor μ represents the trustworthiness of the rule. The generating formula is triggered when the causal relationship occurs in the related fault entity in the fault diagnosis process, and the final rule set is formed through induction.
S4, based on the fault phenomenon of the given current equipment, in combination with the equipment knowledge graph constructed in S2 and the rule set constructed in S3, defining a fault reason intelligent selection technology based on rule matching and knowledge reasoning, screening out a possible fault reason list, and generating a final fault reason, wherein the fault reason intelligent selection technology comprises the following specific steps:
and S401, designing a fault cause intelligent selection technology based on matching and knowledge reasoning based on the equipment knowledge graph constructed in the S2 and the rule set constructed in the S3. For this purpose, firstly, the fault diagnosis rule is designed to be based on the fault phenomenon and fault reason matching method,
Figure BDA0003126611280000081
where s (u, upsilon) represents the similarity between fault phenomena, lυ，iAnd lυ，iAnd respectively matching the fault phenomena upsilon to the fault reasons i and j. And then, training the matching degree of the fault reasons i and j by adopting a random walk model to obtain the matching degree of other similar fault phenomena u to the fault reasons i and j. Based on the process, the matching degree evaluation data of the fault phenomenon is updated in an iterative mode, and a plurality of candidate fault reasons are obtained.
S401, after the matching of the fault reason rules is completed, a plurality of candidate diagnosis reasons are obtained, the final diagnosis reasoning can be converted into a knowledge reasoning problem based on supervised learning, the thought of RankSVM is adopted, a plurality of indexes such as fault equipment, fault phenomena, key elements and the like are comprehensively considered, the partial order relation among the fault reasons is used as a feature vector to be trained, and then the fault reason ordering is converted into a comparative classification problem among the reasons, and the formula is defined as follows:
Figure BDA0003126611280000082
yi(w·(xi-xi))≥1-ξi
wherein the content of the first and second substances,
Figure BDA0003126611280000083
and
Figure BDA0003126611280000084
respectively, the failure cause xiThe scores in the fault features σ and υ, w is the weight vector that needs to be adjusted step by step in the learning process, and the parameter C identifies the degree of tradeoff between model complexity and training error, ξiA non-zero relaxation variable. Training the data based on the optimization function to generate a sequencing model, further finishing sequencing a given fault reason set, and finally selecting the fault reason with the highest rank as a diagnosis result of the current fault.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.

Claims (7)

1. A method for diagnosing equipment fault causes based on knowledge graph and rule constraint is characterized by comprising the following steps:
s1, constructing an equipment fault knowledge graph data model: defining basic elements of equipment fault diagnosis, wherein the basic elements of the equipment fault diagnosis comprise different dimensions of fault phenomena in the equipment fault diagnosis, fault reason elements and semantic relations between the phenomena and reasons; the RDF data model is used for expressing and organizing different dimensions of fault phenomena, fault reasons and semantic relations among the fault phenomena;
s2, constructing an equipment fault knowledge graph: giving a specific device, extracting key elements in fault description, mining the dependency relationship between fault phenomena and reasons, generating a device fault knowledge graph from a fault diagnosis text of an open domain, and representing and storing the device fault knowledge graph by using an RDF (remote data format) graph;
s3, mining the rule from the fault phenomenon set to the fault reason to generate a diagnosis rule constraint set of the given fault;
and S4, based on the fault phenomenon of the given current equipment, defining a fault reason intelligent selection technology based on rule matching and knowledge reasoning by combining the equipment knowledge graph constructed in the S2 and the rule set constructed in the S3, screening out a possible fault reason list, and generating a final fault reason.
2. The method for diagnosing the equipment fault reason based on the knowledge graph and the rule constraint according to claim 1, wherein the step S1 is as follows:
s101, representing equipment fault knowledge graph as binary group
Figure FDA0003126611270000011
Wherein
Figure FDA0003126611270000012
The data mode which represents the basic elements in the equipment fault knowledge graph and describes the equipment and the fault diagnosis process is also called as the equipment fault body,
Figure FDA0003126611270000013
the entity graph is composed of concrete faults, reason factors and phenomena in the equipment fault knowledge graph and semantic relations among the faults, the reason factors and the phenomena;
s102, for equipment fault body
Figure FDA0003126611270000014
Further expressed as doublets
Figure FDA0003126611270000015
Wherein the content of the first and second substances,
Figure FDA0003126611270000016
is a set related to related concepts in the process of equipment and fault diagnosis, and comprises five major classes of equipment names, fault phenomena, fault elements and fault reasons, wherein each major class comprises a plurality of subclasses,
Figure FDA0003126611270000017
is a collection of semantic relationships between concepts, including inheritance, dependency, causal relationships;
s103, pairIn equipment fault knowledge map entity map
Figure FDA0003126611270000018
Can be represented as a doublet
Figure FDA0003126611270000019
Wherein epsilon represents a specific entity corresponding to the equipment fault ontology concept, and comprises a specific equipment, a specific fault phenomenon, a specific fault element and a specific fault reason;
Figure FDA00031266112700000110
is a collection of semantic relationships between equipment entities, where
Figure FDA00031266112700000111
S104, for a given equipment fault scene to be diagnosed, an equipment entity e in the scene belongs to epsilon, and the concept category corresponding to the entity is positioned to an equipment fault body
Figure FDA00031266112700000112
A node of
Figure FDA00031266112700000113
3. The method for diagnosing the equipment fault reason based on the knowledge graph and the rule constraint according to claim 2, wherein the step S2 is as follows:
s201, collecting diagnosis description texts related to equipment faults by taking sentences as units, and expressing the diagnosis description texts as
Figure FDA0003126611270000021
Where n represents the total number of sentences present in the troubleshooting text sequence, using a segmentation tool pair
Figure FDA0003126611270000022
Performing word segmentation to obtain a fault phrase sequence of the equipment
Figure FDA0003126611270000023
m represents the total number of fault phrases, then equipment fault diagnosis experts are organized to label each phrase, and equipment entities are labeled, wherein the equipment entities comprise equipment names, fault phenomena, fault elements and fault reasons, and otherwise, the equipment entities are labeled as non-equipment entities;
s202, defining field characteristics of equipment fault phrases, designing a characteristic selection module, and generating a characteristic vector p corresponding to each phraseiObtaining a feature expression of the whole fault diagnosis phrase sequence
Figure FDA0003126611270000024
Wherein d represents a feature vector dimension for each fault phrase;
s203, giving labeled phrases and feature vectors, wherein labels labeled by equipment fault diagnosis experts in each sentence form a label set whether the phrase is an equipment entity or not, and the label set is represented as
Figure FDA0003126611270000025
Wherein liWhether the phrase is an equipment entity is represented, namely an equipment name, a fault phenomenon, a fault element, a fault reason and a non-equipment entity;
s204, using a classification algorithm to obtain
Figure FDA0003126611270000026
And
Figure FDA0003126611270000027
the discriminant function is obtained from the middle learning
Figure FDA0003126611270000028
It satisfies the minimization function:
Figure FDA0003126611270000029
s205, for the residual texts of the fault diagnosis texts, extracting equipment fault entities of the whole fault diagnosis texts by using the discriminant function;
s206, constructing a relation set between the equipment fault entities based on the semantic relation set defined in the S103;
and S207, representing the equipment fault entity set obtained by mining, the equipment fault entity semantic relation set and the equipment fault body as an RDF (remote data format) graph for storage, and finally generating an equipment fault knowledge graph.
4. The method according to claim 3, wherein the feature selection mode of the feature selection module in step S202 specifically comprises: the equipment fault diagnosis description text specific grammatical features, fault related vocabulary statistical features, phrase semantic features of the equipment fault diagnosis text and phrase structure features of the equipment fault diagnosis text, wherein the fault related vocabulary statistical features comprise field specific words, naming rules and digital text combination specifications, and the phrase semantic features of the equipment fault diagnosis text comprise phrase vector sequences generated by using a pre-training model Bert.
5. The method for diagnosing the equipment fault reason based on the knowledge graph and the rule constraint according to claim 1, wherein the step S3 is as follows:
s301, based on given equipment fault knowledge graph
Figure FDA0003126611270000031
Collecting the existing related fault entities completing fault diagnosis and semantic relations to form a fault diagnosis subgraph;
s302, designing a failure diagnosis causal relationship learning module, and finding a causal relationship between equipment, a failure phenomenon and a failure reason from semantic relationships between different equipment failure entities: firstly, obtaining training data of causal relationship learning from an existing fault-fault phenomenon-fault cause set, wherein a feedforward neural network is adopted to encode the state of each equipment entity in a fault diagnosis subgraph, the relationship between two states is encoded into a real-value matrix form, matrix elements are not fixed {0, 1} binary codes, but attention scores are introduced to serve as confidence degrees of rules to represent the possibility of causal relationship between the entities; secondly, training an LSTM network model, and carrying out modeling representation on the relation and the state of each step;
s303, designing a logic rule generation module, and learning a rule expression for fault reason diagnosis in a fault diagnosis subgraph: firstly, in order to reduce the noise influence of irrelevant fault entities, an attention mechanism is constructed on a fault entity feature vector, a corresponding fault type is judged for a vector with higher weight, further, essential key feature elements for primary fault occurrence are obtained, a rule generator is designed on the basis, and a rule causing the fault occurrence is mined.
6. The method for diagnosing equipment fault cause based on knowledge graph and rule constraint as claimed in claim 5, wherein in step S303, the content of the fault diagnosis rule generator is as follows:
determining a fault entity and a causal relationship set under a given fault, performing feature extraction and logic constraint on key elements related to the fault to form complete logic rule description, introducing a fuzzy factor representing inference credibility on the basis of the definition of an A → B causal relationship production formula to obtain a rule constraint fuzzy production formula in the fault diagnosis process, wherein the rule constraint fuzzy production formula is represented as
Figure FDA0003126611270000032
Wherein the content of the first and second substances,
Figure FDA0003126611270000033
representing a set of rule constraints;
Figure FDA0003126611270000034
representing a fault diagnosis result set corresponding to the rule constraint condition set; the fuzzy factor mu represents the credibility of the rule, and the generating formula is triggered when the causal relationship occurs to the related fault entity in the fault diagnosis process, and the final rule set is formed through induction.
7. The method for diagnosing the equipment fault reason based on the knowledge graph and the rule constraint according to claim 1, wherein the step S4 is as follows:
s401, designing a fault reason intelligent selection technology based on matching and knowledge reasoning based on the equipment knowledge graph constructed in S2 and the rule set constructed in S3, and for this reason, firstly designing a fault phenomenon and fault reason matching method based on fault diagnosis rules:
Figure FDA0003126611270000035
wherein the content of the first and second substances,
Figure FDA0003126611270000041
representing the similarity between the fault phenomena,/v，iAnd lv，iMatching degrees of the fault phenomenon v to the fault reasons i and j are respectively obtained, then, a random walk model is adopted to train the size relation of the matching degrees of the fault reasons i and j to obtain the matching degrees of other similar fault phenomena u to the fault reasons i and j, and based on the process, updating of the matching degree evaluation data of the fault phenomena is iteratively completed to obtain a plurality of candidate fault reasons.
S401, after the fault reason rule matching is completed, a plurality of candidate diagnosis reasons are obtained, the final diagnosis reasoning is converted into a knowledge reasoning problem based on supervised learning, the thought of RankSVM is adopted, fault equipment, fault phenomena and key elements are comprehensively considered, the partial order relation among the fault reasons is used as a feature vector for training, and then the fault reason sequencing is converted into a comparative classification problem among the reasons, wherein the formula is defined as follows:
Figure FDA0003126611270000042
yi(w·(xi-xi))≥1-ξi
wherein the content of the first and second substances,
Figure FDA0003126611270000043
and
Figure FDA0003126611270000044
respectively, the failure cause xiThe scores in the fault features σ and υ, w is the weight vector that needs to be adjusted step by step in the learning process, and the parameter C identifies the degree of tradeoff between model complexity and training error, ξiAnd training the data based on the optimization function to generate a sequencing model, further finishing sequencing a given fault reason set, and finally selecting the fault reason with the highest rank as a diagnosis result of the current fault.
CN202110692310.4A 2021-06-22 2021-06-22 Equipment fault cause diagnosis method based on knowledge graph and rule constraint Active CN113312411B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110692310.4A CN113312411B (en) 2021-06-22 2021-06-22 Equipment fault cause diagnosis method based on knowledge graph and rule constraint

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110692310.4A CN113312411B (en) 2021-06-22 2021-06-22 Equipment fault cause diagnosis method based on knowledge graph and rule constraint

Publications (2)

Publication Number Publication Date
CN113312411A true CN113312411A (en) 2021-08-27
CN113312411B CN113312411B (en) 2023-02-10

Family

ID=77379898

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110692310.4A Active CN113312411B (en) 2021-06-22 2021-06-22 Equipment fault cause diagnosis method based on knowledge graph and rule constraint

Country Status (1)

Country Link
CN (1) CN113312411B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114004360A (en) * 2021-12-30 2022-02-01 北京壬工智能科技有限公司 Intelligent equipment diagnosis equipment and method based on fuzzy expert model
CN114218402A (en) * 2021-12-17 2022-03-22 迈创企业管理服务股份有限公司 Method for recommending computer hardware fault replacement part
CN114723059A (en) * 2022-06-07 2022-07-08 达而观数据(成都)有限公司 Fault attribution system and method based on knowledge reasoning
CN115225457A (en) * 2022-06-24 2022-10-21 南京邮电大学 Knowledge and data dual-drive-based network fault analysis method
CN115508672A (en) * 2022-11-22 2022-12-23 中国电力科学研究院有限公司 Power grid main equipment fault tracing reasoning method, system, equipment and medium
RU2786934C1 (en) * 2021-12-24 2022-12-26 Федеральное государственное бюджетное образовательное учреждение высшего образования "Поволжский государственный университет телекоммуникаций и информатики" Method for predicting the failure of sensor and wireless network equipment based on ontology using machine learning
CN116027673A (en) * 2023-03-29 2023-04-28 中国电子科技集团公司第二十九研究所 Equipment control autonomous decision-making method based on fuzzy neural network
CN116680653A (en) * 2023-06-13 2023-09-01 中国人民解放军海军大连舰艇学院 DS evidence reasoning-based ship-based towed electronic equipment fault field diagnosis method
CN117114102A (en) * 2023-10-13 2023-11-24 江苏前景瑞信科技发展有限公司 Transformer fault diagnosis method based on Bayesian network and fault tree

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108509483A (en) * 2018-01-31 2018-09-07 北京化工大学 The mechanical fault diagnosis construction of knowledge base method of knowledge based collection of illustrative plates
CN109636139A (en) * 2018-11-26 2019-04-16 华南理工大学 A kind of smart machine method for diagnosing faults based on semantic reasoning
CN112307218A (en) * 2020-10-21 2021-02-02 浙江大学 Intelligent power plant typical equipment fault diagnosis knowledge base construction method based on knowledge graph

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108509483A (en) * 2018-01-31 2018-09-07 北京化工大学 The mechanical fault diagnosis construction of knowledge base method of knowledge based collection of illustrative plates
CN109636139A (en) * 2018-11-26 2019-04-16 华南理工大学 A kind of smart machine method for diagnosing faults based on semantic reasoning
CN112307218A (en) * 2020-10-21 2021-02-02 浙江大学 Intelligent power plant typical equipment fault diagnosis knowledge base construction method based on knowledge graph

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114218402A (en) * 2021-12-17 2022-03-22 迈创企业管理服务股份有限公司 Method for recommending computer hardware fault replacement part
CN114218402B (en) * 2021-12-17 2024-05-28 迈创企业管理服务股份有限公司 Method for recommending computer hardware fault replacement parts
RU2786934C1 (en) * 2021-12-24 2022-12-26 Федеральное государственное бюджетное образовательное учреждение высшего образования "Поволжский государственный университет телекоммуникаций и информатики" Method for predicting the failure of sensor and wireless network equipment based on ontology using machine learning
CN114004360B (en) * 2021-12-30 2022-04-01 北京壬工智能科技有限公司 Intelligent equipment diagnosis equipment and method based on fuzzy expert model
CN114004360A (en) * 2021-12-30 2022-02-01 北京壬工智能科技有限公司 Intelligent equipment diagnosis equipment and method based on fuzzy expert model
CN114723059A (en) * 2022-06-07 2022-07-08 达而观数据(成都)有限公司 Fault attribution system and method based on knowledge reasoning
CN115225457B (en) * 2022-06-24 2023-09-26 南京邮电大学 Knowledge and data dual-drive-based network fault analysis method
CN115225457A (en) * 2022-06-24 2022-10-21 南京邮电大学 Knowledge and data dual-drive-based network fault analysis method
CN115508672A (en) * 2022-11-22 2022-12-23 中国电力科学研究院有限公司 Power grid main equipment fault tracing reasoning method, system, equipment and medium
CN116027673A (en) * 2023-03-29 2023-04-28 中国电子科技集团公司第二十九研究所 Equipment control autonomous decision-making method based on fuzzy neural network
CN116027673B (en) * 2023-03-29 2023-06-06 中国电子科技集团公司第二十九研究所 Equipment control autonomous decision-making method based on fuzzy neural network
CN116680653A (en) * 2023-06-13 2023-09-01 中国人民解放军海军大连舰艇学院 DS evidence reasoning-based ship-based towed electronic equipment fault field diagnosis method
CN116680653B (en) * 2023-06-13 2024-01-30 中国人民解放军海军大连舰艇学院 DS evidence reasoning-based ship-based towed electronic equipment fault field diagnosis method
CN117114102A (en) * 2023-10-13 2023-11-24 江苏前景瑞信科技发展有限公司 Transformer fault diagnosis method based on Bayesian network and fault tree

Also Published As

Publication number Publication date
CN113312411B (en) 2023-02-10

Similar Documents

Publication Publication Date Title
CN113312411B (en) Equipment fault cause diagnosis method based on knowledge graph and rule constraint
US20210012199A1 (en) Address information feature extraction method based on deep neural network model
CN109992629B (en) Neural network relation extraction method and system fusing entity type constraints
CN117151220B (en) Entity link and relationship based extraction industry knowledge base system and method
CN111563166A (en) Pre-training model method for mathematical problem classification
CN113312912B (en) Machine reading understanding method for traffic infrastructure detection text
CN114911945A (en) Knowledge graph-based multi-value chain data management auxiliary decision model construction method
CN114969278A (en) Knowledge enhancement graph neural network-based text question-answering model
CN114661905A (en) Power grid fault diagnosis method based on BERT
CN115114455A (en) Ontology-based multi-granularity urban rainstorm waterlogging knowledge map construction method
CN114897167A (en) Method and device for constructing knowledge graph in biological field
Soyalp et al. Improving text classification with transformer
Yu et al. Policy text classification algorithm based on BERT
CN113987201A (en) Zero-sample knowledge graph completion method based on ontology adapter
CN116861269A (en) Multi-source heterogeneous data fusion and analysis method in engineering field
CN115687609A (en) Zero sample relation extraction method based on Prompt multi-template fusion
Qu et al. Knowledge-driven recognition methodology for electricity safety hazard scenarios
CN112380326B (en) Question answer extraction method based on multilayer perception and electronic device
Han et al. A-BPS: automatic business process discovery service using ordered neurons LSTM
CN112632978A (en) End-to-end-based substation multi-event relation extraction method
CN117151222A (en) Domain knowledge guided emergency case entity attribute and relation extraction method thereof, electronic equipment and storage medium
CN117390131A (en) Text emotion classification method for multiple fields
CN115859989A (en) Entity identification method and system based on remote supervision
CN115906846A (en) Document-level named entity identification method based on double-graph hierarchical feature fusion
CN115409122A (en) Method, system, equipment and medium for analyzing concurrent faults of power transformation equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant